Supercomputers Reach The Summit, For Now

Oak Ridge National Laboratory is a facility born from the drive to accomplish scientific feats before an international competitor does the same. In the modern day race for the world’s most powerful computer, the ORNL-housed Summit supercomputer has put the US back in the lead…for now.

BBC:

Summit, the US’s new supercomputer, is more than twice as powerful as the current world leader.

The machine can process 200,000 trillion calculations per second – or 200 petaflops.

China’s Sunway TaihuLight supercomputer, until now the world’s most powerful machine, has a processing power of 93 petaflops.
Summit’s initial uses will include areas of astrophysics, cancer research and systems biology.

It is housed at Oak Ridge National Laboratory (ORNL) in Tennessee, where it was developed in partnership with IBM and NVidia.
Supercomputers are typically large, expensive systems featuring tens of thousands of processors designed to carry out specialised
calculation-intensive tasks.

Summit contains 4,608 compute servers and has more than 10 petabytes of memory

So what does that terminology mean? A quick review from the Knowledge Base, Indiana University:

A 1 petaFLOPS (PFLOPS) computer system is capable of performing one quadrillion (1015) floating-point operations per second. The rate 1 PFLOPS is equivalent to 1,000 TFLOPS. To match what a 1 PFLOPS computer system can do in just one second, you’d have to perform one calculation every second for 31,688,765 years.

A petabyte is equal to one quadrillion (one thousand trillion) bytes, or 1,000 TB. To hold 1 PB of data, you would need about 212,766 single-sided DVDs (a stack that’s about 255.3 meters, or 837.67 feet, tall).

So, what do you do with all that computing power?
Wired:

America’s new best computer is significant for more than just the geopolitics of computational brawn. It’s designed to be more suited than previous supercomputers to running the machine learning techniques popular with tech companies such as Google and Apple.

One reason computers have lately got much better at recognizing our voices and beating us at board games is that researchers discovered that graphics chips could put more power behind an old machine learning technique known as deep neural networks. Facebook recently disclosed that a single AI experiment using billions of Instagram photos occupied hundreds of graphics chips for almost a month.

Summit has nearly 28,000 graphics processors made by Nvidia, alongside more than 9,000 conventional processors from IBM. Such heavy use of graphic chips is unusual for a supercomputer, and it should enable breakthroughs in deploying machine learning on tough scientific problems, says Thomas Zacharia, director of Oak Ridge National Lab. “We set out to build the world’s most powerful supercomputer,” he says, “but it’s also the world’s smartest supercomputer.”

Eliu Huerta, a researcher at the National Center for Supercomputing Applications, at the University of Illinois at Urbana-Champaign, describes Summit’s giant GPU pool as “like a dreamland.” Huerta previously used machine learning on a supercomputer called Blue Waters to detect signs of gravitational waves in data from the LIGO observatory that won its founders the 2017 Nobel Prize in physics. He hopes Summit’s might will help analyze the roughly 15 terabytes of imagery expected to arrive each night from the Large Synoptic Survey Telescope, due to switch on in 2019.

Summit will also be used to apply deep learning to problems in chemistry and biology. Zacharia says it could contribute to an Energy Department project using medical records from 22 million veterans, about a quarter-million of which include full genome sequences.

Impressive as it is, even those responsible for building Summit are looking ahead to what comes next.

Jack Wells of Oak Ridge says the experience of building Summit, which fills an area the size of two tennis courts and carries 4,000 gallons of water a minute through its cooling system to carry away about 13 megawatts of heat, will help inform work on exascale machines, which will require even more impressive infrastructure. Things like Summit’s advanced memory management and the novel, high-bandwidth linkages that connect its chips will be essential for handling the vast amounts of data exascale machines will generate. Scientists at the national lab say they’ve already leveraged Summit’s AI smarts to conduct what is effectively an exascale comparative genomics calculation.

And the competition is never ending, as DOE Secretary Rick Perry noted:

In the most recent chart of the world’s top supercomputers, published in 2017, the US owned 143 of the top 500 while China owned 202.

The US’s previous fastest supercomputer, Titan, was ranked fifth.
“We know we’re in a competition and it matters who gets there first,” said US Secretary of Energy Rick Perry, speaking at the ORNL event.

“The ability to show the rest of the world that America is back in the game and we’re back in the game in a big way is really important.

“Summit’s computing capacity is so powerful that it has the ability to calculate 30 years’ worth of data saved on a desktop in one hour…this is about changing the world.”

What say you? Login and comment.